RISK Analysis
3
American Chemical Society
RISK Education Project
The American Chemical Society (ACS), founded in 1876, is a nonprofit
professional organization and the world’s largest scientific society. ACS
was chartered by Congress in 1937, and its membership includes over
150,000 chemical scientists and chemical engineers.
The mission of the Risk Education Project is to increase the level of
awareness and knowledge in Congress about issues involving risk, includ-
ing assessment, management, characterization, policy, and communica-
tion. The project is funded by a generous grant from the Eastman Kodak
Company.
4
Contents
1) Why the Attention to Risk Analysis? 6
The Emergence of Risk Analysis in Health, Safety, and
Environmental Policy
Further Reading 38
5
1) Why the Attention
toRISKAnalysis?
6
chapter one
assessment. From the 1930s onward, professional organizations such as the
American Society of Mechanical Engineers (ASME) and the American
Conference of Governmental Industrial Hygienists (ACGIH) used this
knowledge base to establish various health and safety codes and standards per-
taining to industrial equipment, activities, and exposures to noxious agents.
When concerns turned to low-dose exposures to ionizing radiation and
to potentially carcinogenic chemicals in the 1960s, the NOEL approach to
hazard management proved problematic, because no-effect threshold levels
could not be established. This shortcoming prompted considerable basic
scientific research to understand the effects of low-dose exposures on
humans. In the 1960s, the Nuclear Regulatory Commission (NRC) began
to use computer models to estimate the human risks of exposures to ioniz-
ing radiation. And in the early 1970s, the Food and Drug Administration
(FDA) began to employ similar methods in its evaluation and regulation of
potential carcinogens in foods.
Congress’s historic legislation of the early 1970s, establishing a statutory
framework for regulating health, safety, and environmental hazards, and the
creation of the Environmental Protection Agency (EPA) and the
Occupational Safety and Health Administration (OSHA) considerably raised
the role of risk analysis in the regulatory process and stimulated profession-
alization of the field. In the intervening decades, risk analysis has become
an increasingly central component of the nation’s policy making concerning
health, safety, and environmental quality. Modern, quantitative methods of
contemporary risk assessment emerged in the mid-1970s, and they have
been applied to a wide variety of regulatory issues, including pesticide
residues; food additives; pharmaceutical agents; pollutants in drinking water,
soil, ambient air, indoor air, and other environmental media; the safety-
related features of industrial processes and transportation equipment; and
other hazardous aspects of consumer products.
Most observers would agree that over the past several decades the
advancement of risk analysis in regulatory decision making has promoted
rational policy deliberation. Yet, as real-world practice indicates, risk analy-
ses have often been as much the source of controversy in regulatory consid-
erations as the facilitator of consensus. The current state of scientific
understanding has often been found to be incomplete, indecisive, and con-
troversial in attempting to resolve the important questions about the type
and size of specific hazards. Additionally, considerations in risk manage-
ment—issues of risk acceptability and how to balance trade-offs among
competing interests—are beyond the technical/scientific debate. Thus, in
deciding what we might expect from a policy-making process in which risk
analysis is prominent, it is essential to recognize the character, strengths,
and limitations of the analytical methods involved.
7
Analyzing, Managing,
)
2RISKand Communicating
An Overview
RISK Assessment
Risk assessments are conducted to estimate how much damage or injury
can be expected from exposures to a given risk agent and to assist in
judging whether these consequences are great enough to require increased
management or regulation. Depending on the kind of hazard, the effects
of primary concern might be workplace injuries; reproductive and genetic
abnormalities; diseases such as cancer or other debilitating illnesses; or
ecological effects such as species extinction, loss of habitat, and other
kinds of ecosystem damage.
In the health, safety, and environmental fields, risk is usually identified
as the likelihood that individuals (or a population) will incur an increased
incidence of adverse effects such as disabling injury, disease, or death.
Risk frequently is expressed in quantitative probability terms—such as
some number of additional cancer deaths over a lifetime in a population of
1 million exposed people. (A risk of 1 in 10,000 is often described as a
“10–4 risk”, 1 in 1 million as a “10–6 risk”, and so on.) Historically, risks
of less than 10–6 in magnitude have not been the object of concern.
More qualitative characterizations are also used—such as low, medium,
high—where quantification is either infeasible or unnecessary.
Risk assessments range widely in scope and complexity, depending on the
application: from simple screening analyses to major analytical efforts that
require years of effort and a substantial budget. Contemporary risk assess-
ments ordinarily rely on many branches of science—on the methods and
knowledge of disciplines such as toxicology, epidemiology, other health and
environmental sciences, systems engineering, and related technical areas.
8
The methods and sequence of steps involved in conducting a risk
assessment vary with the kind of risk and its possible consequences. A more
specific discussion of these elements for several key risk assessment areas
follows in a later section. In its most general form, however, the process
consists of a source assessment, an exposure assessment, an effects assessment,
and is normally concluded by an integrative risk characterization.
chapter two
Source assessment seeks to identify and evaluate the sequences of events
through which an exposure to a risk agent could arise. In risk assessments
of engineering systems, for example, this can be a particularly extensive
and detailed exercise—such as evaluating the possibility that a pump in a
manufacturing operation might fail, leading through a series of steps to
increased levels of toxic substances on the shop floor. Alternatively, this
kind of analysis might be aimed at finished products, whose physical fea-
tures along with typical use patterns could result in safety hazards.
Exposure assessment seeks to determine the number and kinds of people
exposed to a risk agent, along with the magnitude, duration, and timing
of their exposures. An example is estimating the fate and distribution of a
toxic chemical released from a manufacturing facility and providing a
description of the characteristics of the exposure of human populations
along the path of the chemical. Depending on the needs of the analysis,
the evaluation might focus on current, past, or future exposures.
Effects assessment determines the extent of adverse effects likely to result
from given levels of exposure to a risk agent. For resource and efficiency
reasons, this kind of analysis is usually conducted in stages. The initial
analytical step is to determine if exposures to a risk agent at any level
could cause adverse effects—for example, whether exposures to a particu-
lar industrial chemical could cause cancer or seriously impair nervous sys-
tem function. Then, if such a conclusion is drawn, a more detailed study
is conducted to determine what quantitative relationship (dose–response)
exists between the level of exposure and the incidence of adverse effects.
Risk characterization is the concluding step of a risk assessment. This
is an important integrative task, which involves assembling the prior analy-
sis components into a bottom-line picture of the nature and extent of the
risk. The principal topics include the kinds of health effects likely to arise,
the risk’s potency (i.e., the severity of the adverse effects), the populations
affected, the likelihood of exposure, and the risk’s ultimate magnitude
(i.e., potency adjusted for the likelihood of exposure). Risk characteriza-
tions are usually the principal means through which a risk assessment’s
findings are communicated to risk managers, policy makers, journalists,
and the public.
In the past, risk characterizations have frequently consisted of brief
descriptions of potential adverse effects and affected populations, along
9
with a single numerical estimate of the level of risk that would summarize
whether humans would experience any of the various forms of toxicity or
other effects associated with the risk agent. (Often this figure has been in
the form of a plausible upper bound on risk, deliberately prepared to pro-
vide a conservative estimate that minimizes the chance of underreporting
the actual level of risk.) More recently, however, this “short form”
approach to risk characterization has been criticized. It is now generally
acknowledged that characterizations need to provide deeper insight into
how risk estimates and findings are generated (including a discussion of
the assumptions that underlie the calculations). In addition, characteriza-
tions should consider a range of plausible risk estimates (which could
result from the use of plausible alternative assumptions or differing models
of exposure and dose–response relationships) and should more clearly dis-
cuss the uncertainties and limitations in the empirical data on which the
risk assessment is based.
10
RISK Management
The essential tasks of risk management are to (1) determine what hazards
present more danger than society (as represented by its government) is
willing to accept; (2) consider what control options are available; and (3)
decide on appropriate actions to reduce (or eliminate) unacceptable risks.
chapter two
At the broadest level, risk management includes a range of management
and policy-making activities: agenda setting, risk reduction decision mak-
ing, program implementation, and outcome evaluation. As discussed in a
later section, the nation’s laws and policy programs directed at risk are
made up of a complex framework through which decisions about risk
management are made.
Risk assessments provide a basic input to risk management. However,
such assessments do not of themselves provide answers to many questions
that risk managers must answer. What level of exposure to a risk agent is
an unacceptable risk—and, conversely, what level is acceptably safe? How
should uncertainties about the extent of risks be hedged? What trade-
offs should be made among risk reduction, benefits derived, and new
costs incurred in achieving improved risk control? Will new risks arise as
a consequence of reducing existing risks—and how should such trade-offs
be considered should they arise? Which of the existing hazards deserve
the greatest attention and resources? Such issues are clearly influenced
by society’s values and priorities, and dealing with them requires the
political considerations associated with establishing policy in a democratic
manner.
There are several policy approaches to hazard reduction. Command
and control measures, which include regulations, permits, and enforce-
ment actions, represent one avenue that has a long-standing history in
U.S. risk management. Other options include market-based economic
incentives that prompt desired changes in industrial production decisions
and consumer behaviors; voluntary reductions of risk-producing activities;
promotion of pollution prevention; and information and education pro-
grams to modify behaviors by alerting consumers and technology users of
the risk involved in their choices.
Ultimately, those with the responsibility to manage risks in society’s
interest must make responsible decisions about risk control and hazard
reduction—and work through issues that are not always easy to resolve,
including legal obligations, uncertainties in the risk assessment evidence,
and trade-offs among competing interests to protect the public’s health
and welfare. Risk assessment, cost analyses, and other analytical tools
can assist the good judgment of the policy maker in making such deci-
sions.
11
RISK Communication
Risk communication covers a range of activities directed at increasing the
public’s knowledge of risk issues and participation in risk management.
This includes, for example, warning labels that provide consumer educa-
tion about existing hazards, development of publicly accessible databases
characterizing hazardous circumstances, and public hearings on risk man-
agement issues.
Risk communication emerged as a recognized element of risk manage-
ment early in the 1980s. At this time, it was realized that a large fraction
of the public was not familiar with the nature of risk and that risk manage-
ment decisions could not simply be made by technical experts and public
officials and then imposed and justified to the public after the fact. Risk
communication is now viewed as being a dialogue among interested par-
ties—risk experts, policy makers, and affected segments of the public
12
3) RISK
in
Analysis
Regulatory
Decision Making
Agency Conduct of RISK Analysis
In the United States, over the last several decades, risk assessments
have become increasingly prominent inputs to the standard-setting
and program implementation efforts of such federal agencies as the
Consumer Product Safety Commission (CPSC), the EPA, the FDA, the
chapter three
NRC, OSHA, the Department of Agriculture, and the Department of
Transportation, as well as state-level agencies and offices with similar
responsibilities.
Decision makers responsible for managing risks now rely on risk assess-
ments to estimate which of the existing hazards are significant enough to
warrant policy attention. The scope of the risk assessments employed
varies from case to case—ranging from relatively simple screening assess-
ments using standard assumptions about exposures, to much more expan-
sive studies involving data collection and modeling, to highly detailed
analyses of situation- or site-specific circumstances.
13
laws that illustrate these differences. Most fall under one of the following
three categories:
• Health-based standards. For health-based standards, the mandate
requires hazards to be regulated without regard to cost factors or the cur-
rent availability of suitable control technology. For example, Section 109
of the 1970 Clean Air Act directed EPA to establish standards for certain
air pollutants that provided an “ample margin of safety” for the “most
sensitive” groups. The standards were to be set based only on whether
health risks existed and regardless of new costs imposed or technological
limitations.
• Technology-based standards. In this case, the mandate requires the
adoption of “best practicable control technology,” “best available technol-
ogy,” or other similar kinds of pollution controls or treatments. Here, the
overriding considerations are not risk reduction but the cost and efficacy
of a control measure in reducing pollutant or contaminant concerns. An
example is the Safe Drinking Water Act, where maximum contaminant
level goals are specified based solely on health considerations, but the
actual standards are developed on technological feasibility and cost
grounds.
• Risk-balancing standards. Here, the balance of the benefits of risk
reduction against the costs incurred is considered in setting risk manage-
ment goals. Under the Federal Insecticide, Fungicide, and Rodenticide
Act, EPA’s regulation of pesticides must seek to balance the health and
environmental impact of a chemical, the costs of regulation, benefits, and
other societal concerns.
14
TABLE 1
Representative Risk-Related Provisions in Selected Federal Statutes
Occupational Safety
and Health Act
Sections 3(8) Safety and health Material impairment of Attain highest degree
and 6(b)(5) risks in the work- health or functional capaci- of health and safety
place ty protection
What is reasonably neces- Best available evidence
sary or appropriate to pro- Technical and economic
vide safe and healthful feasibility
employment
Clean Air Act
Section 109 Set standards to provide
National Ambient Air Protect public health ample margin of safety
Quality Standards
Section 112 Reduce emissions using maxi-
chapter three
Emissions standards Adverse effects to health mum achievable control
for hazardous air and the environment technology, and later address
pollutants residual risk
Toxic Substances
Control Act
Section 6 Balance risks against econom-
Existing chemicals in Unreasonable risks to health ic benefits, considering alter-
commerce and the environment native technologies
Federal Insecticide,
Fungicide, and
Rodenticide Act
Section 3 Balance risks against econom-
Pesticides Unreasonable risks to health ic benefits to pesticide users
and the environment and society
Comprehensive
Environmnetal
Response,
Compensation , and
Liability Act
Section 303 Reporting is based largely on
Toxic Release Hazards to human health or hazard and quantity used
Section 9621 Inventory the environment Protect human health and the
Hazardous waste site Persistence, toxicity, mobili- environment in cost-effective
remediation ty, and propensity to bioac- manner
cumulate, short- and long-
term health effects
Safe Drinking Water
Act
Section 300g-1(b) Set a goal (maximum contam-
Drinking water quali- Known or anticipated inant level goal, MCLG)
ty adverse effects on human with an adequate margin of
health safety and define a maxi-
mum contaminant level
(MCL) as close as feasible to
the goal
Source: Office of Science and Technology Policy, Executive Office of the President, Science, Risk, and
Public Policy, March 1995.
15
Many statutes mix the elements of the above categories. For example,
OSHA’s setting of health standards under the Occupational Safety and
Health Act requires that where the risk of health impairment is “signifi-
cant” (such as workplace exposures to carcinogenic or other toxic sub-
stances) the responsible substances are to be fully removed (via perfor-
mance-based standards) from the workplace—constrained only by “techno-
logical and economic feasibility”.
16
tions that are worth their cost. Critics of this approach find that the very
process of risk assessment allows some level of risk to be considered
acceptable (i.e., that a risk–benefit balance can be found); this may be
unlawful under some statutes and may be perceived as unethical in some
circumstances. Others conclude that risk assessments are a tenuous basis
for policy making because of the data and knowledge limitations and
other uncertainties frequently encountered in conducting them.
Other risk management approaches that have been used in the United
States and other nations include mandating reductions of risks to levels
that are as low as reasonably achievable (ALARA), requiring adoption of
best available technologies (BAT), or imposing outright bans on the
use/consumption of hazardous agents. These alternative approaches,
however, have their shortcomings as well. ALARA or BAT can be costly
to implement and may not result in a worthwhile benefit to society—if,
chapter three
for example, the magnitude of the uncontrolled risk is not high. Likewise,
banning a risk agent may not guarantee a significant risk reduction or jus-
tify the associated financial sacrifice, because, for example, a ban could
result in the use of a hazardous substitute.
17
4) RISKAssessment
Methods
Hazardous agents and sources can differ consider-
ably with respect to the phenomena gov-
erning their occurrence and the kinds of effects produced in exposed pop-
ulations. As a general rule, the approaches and methods for conducting
risk assessments must be tailored to match these features—which makes
risk assessment far from a “turn the crank” analytical process. In health,
safety, and environmental policy issues, the tools of health risk assessment,
engineering systems risk assessment, and ecological risk assessment pro-
vide the essential foundation for risk analysis. The basic features of each
of these fields are discussed below.
18
coordinate the various risk analysis practices that had come to be
employed, recommended a four-step process for conducting health risk
assessments. This process, outlined below, has become the standard
model for the field.
Hazard identification. This initial risk assessment activity is directed at
determining if a substance (or other health-threatening risk agent) could
cause particular adverse health effects in human populations. For exam-
ple, will exposure to a particular substance cause cancer? Will it harm the
nervous system or immune system? Will it give rise to reproductive
defects or other serious health conditions or disabilities?
Dose–response assessment. This step seeks to identify the quantitative
relationship between a dose level and the resulting incidence of injury or
disease. Most substances—even many of those used for beneficial purpos-
es—cause harm when consumed in large enough quantity. For example,
an anesthetic may cause headaches at low doses, a medically advantageous
sleep at higher doses, but is lethal at very high doses. Thus, the riskiness
of a substance cannot be determined with confidence unless the
dose–response relationship is quantified.
With noncarcinogens (or the noncarcinogenic effects of carcinogens),
the normal working assumption (backed up by theory and empirical evi-
dence) is that biological effects occur only after a threshold level of expo-
sure has been exceeded. Various thresholds have come to be established;
chapter four
they include a lowest observable effect level (LOEL), the smallest dose that
causes any detectable effect; a no-observed-effect level (NOEL), the dose at
or below which no biological effects of any type are detected; and a no-
observed-adverse-effect level (NOAEL), the dose at or below which no
harmful effects are detected. Toxicologists generally seek to identify (via
animal testing, with progressively higher and lower exposure levels to a
suspected toxic substance) several of these dose–response markers to help
map thresholds.
With suspected carcinogens, however, the working assumption is usual-
ly that no threshold exists, (i.e., that exposures to carcinogenic substances
pose some risk even at the smallest level of exposure). This concept has
long been a mainstay of cancer risk assessment; the concept is based pri-
marily on what is known about the health effects mechanisms associated
with exposures to ionizing radiation and toxic substances. More recent
findings about the induction of cancer suggest that a wider range of
mechanisms may be at play, depending on the substance involved, and
that for some carcinogens there may be a threshold below which cancer
does not occur.
Exposure assessment. This step attempts to identify the nature and size
of the population(s) exposed to the risk agent, along with the magnitude,
19
duration, and spatial extent of exposure. Depending on the purpose, the
exposure assessment could concern past or current exposures or those
anticipated in the future.
Case by case, the steps involved in an exposure assessment vary widely,
because circumstances differ with respect to how much is known about
existing exposures and what information is available. The most reliable
picture comes from direct monitoring (personal, biological, and/or ambi-
ent) of the amounts of the substance to which people are actually exposed
over time. This sort of information is, however, often not available. In
fact, lack of knowledge about actual exposures is one of the weaker links
in the knowledge chain supporting risk assessments. As a consequence, a
good deal of what is done is derived from models and from generalized
assumptions about relevant physical parameters and human behaviors.
Numerous pathways exist through which exposures can occur (direct
and indirect), and a large number of variables and moderating factors can
be involved. For example, estimating the movement of a chemical in the
environment depends on considerations such as how easily it evaporates,
how easily it dissolves in different media (such as water or animal fat),
how strongly it attaches to the soil, and how long it persists in the envi-
ronment. On the human behavior side of the exposure equation, the
issues include how much water or specific types of food people consume
each day; whether or not people filter their water; how they prepare their
food; what balance of time during the day is spent indoors versus out-
doors, and so on.
Risk characterization. This concluding task in a risk assessment com-
bines the principal findings of the hazard characterization, dose–response,
and exposure phases of the risk assessment into an integrated picture of
the nature and expected frequency of adverse health effects in exposed
populations. Ordinarily, the “bottom line” forthcoming from a risk char-
acterization is a primary determinant of the risk management phase that
follows risk assessment.
20
and other relevant substances can be well documented, the exposed popu-
lation well defined, and the kinds of adverse effect(s) known in advance,
epidemiology provides the most direct way of determining the effects of a
risk agent on human biology.
Epidemiology’s weakness is that these essential conditions are often not
met. The presence of confounding factors, such as simultaneous expo-
sures to other toxic substances, can make the health effects of a given risk
agent difficult to determine.
Additionally, it is difficult to accurately account for population mobility
and the genetic variability of humans (which can, among other things, sig-
nificantly affect an individual’s susceptibility to many diseases). Moreover,
epidemiologic studies are usually not sufficiently sensitive (in a statistical
sense) to allow the detection of small changes in risk levels (such as might
be associated with low-dose exposures to chemicals in the environment).
• Toxicological studies. Most of the information used to predict the
adverse health effects of exposures to substances comes from animal test-
ing or test tube procedures using cells or tissues isolated from animals or
humans. These kinds of studies allow the examination of potentially toxic
substances while accounting for different exposure levels and genetic vari-
ability. Animals, including the rodents frequently used in toxicological
testing, biologically resemble humans in many ways. A good body of evi-
dence indicates that animal studies can be used in many (but not all)
chapter four
instances to deduce hazards to human health—although, not always to
indicate the precise level of risk that humans would face.
Considerable research has been conducted over the years in toxicology
and has been directed at developing and employing various animal models
to predict adverse health effects in humans; understanding the mecha-
nisms of toxicity; and assessing the extent to which biological processes
and toxic effects are similar in test animals and humans. One of the par-
ticularly significant advances over the last decade is the development of
physiologically based pharmacokinetic (PB-PK) models as a basis for pre-
dicting human health effects from rodent data. These models seek to
account for the various differences between test species and humans by
considering body weight, metabolic capacity and products, respiration
rate, blood flow, fat content, and other biological parameters.
Despite their merits, toxicological studies suffer from some serious lim-
itations. Cost considerations typically limit the number of test animals to
a few hundred. To make up for such limited sample sizes, research
designs must use high exposure levels—perhaps 1000 times or more
greater than typical human environmental exposures—to maximize the
likelihood that health effects can be detected with acceptable statistical
21
precision. (Without such high exposures, study designs might require
millions of experimental animals.) As a consequence, estimating a sub-
stance’s ability to cause adverse health effects at the (low) levels typical of
environmental exposures depends on extrapolating the dose–response rela-
tionships from the (high exposure) experimental data down to levels well
below the range verifiable by experimental data.
A number of mathematical models, based on various scientific theories
of how toxic substances cause biological effects, have been developed to
perform such extrapolations. The scientific community, however, often
disagrees on what is the most appropriate theory. In addition, the models
associated with differing health effects theories can give vastly different
estimates of the level of risk associated with human exposures (see discus-
sion later in this section).
• Structure–activity studies. This kind of analysis seeks to evaluate toxi-
city based on the substance’s chemical structure. The large and still grow-
ing base of empirical knowledge about molecular structure and toxicity
has made this approach more feasible.
Nevertheless, such relationships are not simple. Experience has shown
that, although this method can be informative, its predictive capacity is far
from perfect.
• Exposure data and exposure modeling. Information about the expo-
sures of human populations to risk agents is a crucial input to the risk
assessment process. Risk assessors need this information to estimate the
amount of the substance that reaches the cells, tissues, or organs of
exposed individuals. In general, exposure assessment involves identifying
the pathways (e.g., air, food, and water) by which a substance travels
through the environment; the changes the substance undergoes en route;
its environmental concentrations relative to time, distance, and direction
from its source; the routes through which human exposures could occur
(e.g., oral, dermal, and inhalation); and the distribution of sensitive pop-
ulation subgroups (such as children or pregnant women).
When reliable data on actual human exposures are not available—a not
infrequent circumstance—the gaps in information must be filled in by
simulation models, generalized assumptions, or some combination of the
two. To provide estimates of the exposure levels experienced, information
on the movement and activity patterns of at-risk human populations is
coupled with modeled estimates of the transport and distribution of sub-
stances from their sources to various environmental media (the atmos-
phere, ground or surface water, soil, the food chain).
Various computerized simulation models have been developed to esti-
mate the transport and distribution of substances in the environment.
22
Some are specific for classes of substances whereas others are based on the
various media in which transport occurs. Still others are multimedia in
nature and address the combined impact of numerous routes of exposure.
Many risk assessments draw on several or all of these types of evidence.
Additionally, other kinds of studies can play important roles. For exam-
ple, basic research on metabolism, pharmacokinetics, and the mechanisms
of toxicity is often used to evaluate the relevance of the above approaches
in predicting adverse health effects in humans.
Sources of Controversy in
Health RISK Assessments
The risk assessment process as outlined above has a number of strengths:
a structure for collecting, organizing, and evaluating data; a capacity to
base policy decisions on the estimated level of human risks; a basis for
focusing research efforts on important risk assessment topics; and, in prin-
ciple, a basis for ranking risks and focusing hazard management resources.
Nonetheless, the process has a number of limitations: It can involve
exceedingly complex analyses, with much judgmental weighing of diverse
data; it is vulnerable to limitations in data and to uncertainties in scientific
reasoning; and it requires a good many assumptions, at least some of
which will be debatable. Risk assessment findings are frequently contro-
versial, for one or more of the following reasons:
chapter four
• Risk inferences dependent on animal testing. Human data provide the
most straightforward basis for gauging the level of human health risk. For
various (and usually compelling) reasons, animal bioassays remain a central
risk assessment tool, particularly in assessing possible carcinogens. Animal
testing is widely regarded as the next best approach, if adequate human
data are not available. Supporters of animal testing conclude that it pro-
vides a reasonable basis for assessment.
There are problems, however: (1) The high (in terms of level or dura-
tion) exposures used in the standard animal test designs usually have no
parallel in humans, thus creating the need for extrapolations to levels out-
side that verifiable by experimental data (a situation science shies from).
(2) The high exposures may provide a misleading picture of the potential
for health effects, because it is possible that the high doses induce effects
that do not arise at lower doses. (3) Finally, some toxic mechanisms and
pathways that occur in animals may not occur in humans.
Although animal models may be useful in many cases, there is always a
chance of false negatives or false positives. One particularly graphic exam-
ple of this circumstance is the drug thalidomide (prescribed during preg-
nancy for therapeutic reasons), which caused severe birth defects in the
23
children of women taking the substance. However, no adverse health
effects were found in animal testing.
When a dose–response relationship derives from data on test animals,
the need for an additional extrapolation arises: namely, translating the
observed effects in animals to predicted risk in human populations. The
usual approach is to introduce one or more inter-species extrapolation
adjustments (usually termed scaling factors) to account for biological dif-
ferences between the test animal and humans (such factors as lifespan,
body size, genetic variability, and metabolic rate, among others) that may
influence the response to exposures of a given substance. Working con-
ventions about these adjustments have emerged over the years, but their
exact nature remains controversial among scientists.
In principle, the human response to the substance in question may be
more or less sensitive than what is observed in the animal tested. None-
theless, the working convention has been to assume greater sensitivity on
the part of humans. Numerically speaking, the risk estimates from test
animal dose–response data are substantially adjusted upward by 1 or more
powers of 10. In a few cases, epidemiologic data are available to gauge
the dose–response relationship. However, for most cases, the low level of
most environmental exposures to suspected carcinogens combined with
the lack of a threshold means that estimates of human cancer risk rely on
the low-dose predictions from mathematical extrapolation models. The
extrapolations are based on comparatively high dose levels given to labora-
tory animals.
Presently, about half a dozen extrapolation models are used routinely
(including the one-hit, multistage, multihit, Weibull, and probit models).
Many federal regulatory agencies have traditionally employed the lin-
earized, multistage model, which does not exhibit a threshold and is based
on the concepts that cancer develops in stages and that a carcinogen can
have an effect at each stage. The analytical difficulty is that different mod-
els may all fit the (high-dose) experimental data extremely well but yield
starkly different predictions of risk at the low exposure levels that are of
primary concern for real-world risk management decisions (see Chart 1).
Recent research has tended to indicate that some carcinogenic threshholds
exist, and therefore other models may prove more appropriate.
• Weight of evidence. Ordinarily, risk assessors will review and consider
the findings of numerous published studies in the process of conducting a
risk assessment. Such material often varies considerably in quality. In the
past, regulatory agencies often placed heavy emphasis on any study (regard-
less of quality) that showed a substance to be a potential health hazard.
This strategy, which minimizes the potential of underestimating the actual
24
CHART 1
Low-Dose Extrapolations in Cancer Risk Assessments—
the Challenge of Predicting the Level of Human Risk
Estimating the level of human cancer risk depends on extrapolating the
health effects observed in the high-dose experimental data with laboratory
animals down to the much lower levels typical of environmental exposures.
Various models have been developed for this purpose, reflecting differing
concepts of the biological mechanisms of cancer induction at work.
The difficulty of this approach, however, as well illustrated below by a
risk analysis for DDT, is that differing extrapolation models that fit experi-
mental data well can yield vastly different estimates of the level of risk at
low exposure levels. On the basis of current scientific knowledge, it is
often not easy (or possible), to conclude which of the models provides the
most reliable estimate.
In this evaluation of the cancer risk of DDT exposure, the output of
each of the three extrapolation models considered (all widely used in risk
assessment work) fits the experimental data well and provides similar esti-
mates of the level of risk across much of the observed exposure range.
However, the models’ predictions of risk levels become increasingly dis-
parate at the lower dose levels more characteristic of human environmen-
tal exposures. At the smallest level shown here, the predicted levels of risk
span a difference of nearly 4 orders of magnitude—and range from a level
chapter four
that would be of clear public health concern (10–4) to one that probably
would not (10–8).
10–4
Response rate
Response rate
60
30
10–8
Multihit
Weibull
10 Multistage
25
level of human risk in the face of uncertainty, has provoked much contro-
versy.
This problem began to recede in the 1990s, when the scientific com-
munity and most regulators accepted that not all data should be given
equal status, and that only data of similar quality should receive equal
weight. This weight of evidence approach takes into consideration the
quality and adequacy of the available studies (in hazard identification,
dose–response, or exposure assessment).
• Conservative bias in assumptions. Assumptions invariably play a sig-
nificant role in the quantitative stages of risk assessments—filling in where
hard data may not be readily available and providing a way forward where
scientific understanding may be incomplete or nonexistent. Regulatory
agencies have developed a host of standard assumptions (often termed
defaults) for use in these circumstances (and when evidence to the con-
trary is absent).
The 1983 National Academy of Sciences report (Risk Assessment in the
Federal Government: Managing the Process) referred to 50 previously
identified points in the risk assessment process where scientific uncertainty
is encountered and assumed links are needed to make predictions. These
circumstances span most of the critical analytical areas in a risk assessment:
translating animal risk estimates to human risk predictions, dose–response
extrapolations, and various aspects of exposure assessment.
Critics have argued that these defaults tend to be very conservative—
erring excessively on the side of caution and resulting in predictions that
are likely to overstate the actual level of human risk. Examples include the
use of test data from the most sensitive species, the choice of a dose–
response extrapolation model that yields the highest estimation of risk
(usually, the linearized, multistage model), and exposure scenarios that
focus on maximally exposed individuals rather than that typical for the
majority of individuals in the exposed population.
Regulatory agencies defend their choices as prudent, given the substan-
tial scientific uncertainties and their statutory responsibility to protect the
nation’s health (which may specify protection of the most susceptible indi-
viduals in exposed populations). Furthermore, some analysts believe that
even with these conservative assumptions, risks may still be underestimat-
ed. For example, test animals are not exposed at the beginning of their
lives when they are more susceptible to some substances, and the cumula-
tive effect of exposure to a substance from many possible routes or multi-
ple substances is usually not considered in risk assessments.
The debate on this topic goes on—and will not soon end. There is
now greater recognition that risk characterizations should provide a better
26
account of the full range of credible risk estimates supported by the evi-
dence (i.e., not just the upper-bound estimate forthcoming from the
model that yields the highest risk prediction) and the significance of other
alternative assumptions in the risk calculations (such as with regard to
exposures).
• Human variability. In most cases, substantial variation will exist
among individuals in their susceptibility to a given substance. Small chil-
dren, for example, may be more vulnerable than adults. Additionally,
genetic variability in humans can give rise to substantial differences across
a population. Although both of these considerations are important to the
estimation of risk, they add considerable difficulty to the assessment
process.
• Mixtures and interactions. Day-to-day life involves exposures to a
large number of substances, either concurrently or sequentially, by various
routes, from a variety of sources over varying periods of time. None-
heless, with very few exceptions, risk assessments and the regulations that
result have dealt with a single type of exposure to a single chemical. Little
is known about the effects of most chemicals when encountered in mix-
tures.
The simplest and most common assumption employed is that the risk
characteristics of the components of a mixture are straightforwardly addi-
tive. However, the combined effect may be either synergistic (in which
chapter four
the overall risk of the mixture is greater than predicted by the additive
model) or antagonistic (in which the overall risk is less than predicted by
the additive model). For example, studies have shown that simultaneous
exposures to tobacco smoke and radon result in a health risk that is
greater than posed by exposure to either alone. Alternatively, some evi-
dence shows that dioxin, when present before exposure to another car-
cinogen, works to reduce the rate of cancer from exposure to the second
carcinogen.
• Treatment of uncertainties. A good many risk assessments—particu-
larly those where carcinogenicity is suspected—do not reduce to a single
risk figure or characterization that can be cited with confidence. The exact
relationship of animal tests to human risk and the predictive value of high-
exposure occupational epidemiology to environmental exposures are
unclear. It is difficult, in many cases, to determine the most appropriate
mathematical model for extrapolating from high to low doses in
dose–response evaluation. Additionally, exposure assessments (particularly
where many pathways are involved) are often substantially simplified rela-
tive to the real-world circumstances.
Advances in science will help to diminish these limitations. However,
27
debatable assumptions, plausible risk estimates that contrast sharply, and var-
ious other uncertainties are likely to remain the name of the game in health
risk assessment for some time to come—making risk management an exer-
cise conducted under uncertainty. The risk assessment community now rec-
ognizes that risk managers need better understanding of the uncertainties in
risk analyses, along with insight about the sensitivity of the findings to possi-
ble alternative assumptions (regarding dose–response assumptions, scaling
factors, exposure scenarios, etc.). Risk characterizations, in addition to pro-
viding a quantitative estimate of risk or a range of possible values, should
also discuss the assumptions involved in determining the magnitude of risk,
the strengths and weaknesses of the evidence used, the significance of any
uncertainties that remain, and the implications of any probable alternative
assumptions that might have been made in risk calculations.
A second (and reinforcing) development is the use of modern mathe-
matical/probabilistic techniques in risk assessment, which facilitates the
treatment of uncertainties in risk calculations.
28
Ecological RISK Assessment
Even though there is a long history of evaluating environmental and
ecosystem impacts, the concept of ecological risk has only recently
emerged as a distinct field of risk assessment. Ecological risk is based on
the understanding that healthy ecosystems can provide renewable
resources and food, water storage and flood control, biodegradation and
removal of contaminants from air and water, pest and disease control,
moderation of climatic extremes, recreational opportunities, and scenic
beauty. The objective of an ecological risk assessment is to estimate the
possibility of adverse impacts on one or more of these ecosystem dimen-
sions from exposures to ecosystem/environmental stressors such as tech-
nology (roads, buildings, and other types of development) and pollutants.
The scope of an ecological risk assessment may be fairly narrowly defined,
such as the adverse effects of development in a local wetlands area, or
widely encompassing, such as the worldwide issues of global climate
change.
Ecological risk assessment is increasingly viewed as relevant to complex
ecological problems and policy questions, such as the decline of Pacific
salmon in the U.S. Pacific Northwest or the current decrease in biological
diversity in many parts of the globe. To date, however, applications of
ecological risk assessment have focused chiefly on the risks of chemicals in
chapter four
the environment, with the impacts on animal species as surrogates for
measuring ecological health.
Methodologically, ecological risk assessment draws widely on the stan-
dard procedures of environmental impact analysis and monitoring. Tests
of animal species (such as fish or insects) are commonly used tools as are
computer-assisted geographic analysis and computerized ecosystem simu-
lations. Additionally, expert judgments and opinions also play a substan-
tial role because ecosystems are complex and do not lend themselves
entirely to experiments or modeling.
Ecological risk assessment’s greatest challenge is to work out practical
concepts and measures of what comprises ecological health at a whole sys-
tem level. At the root, this involves deciding what ecological changes are
adverse (and, thereby, undesirable) and which are beneficial (and desired).
Scientific theory goes only so far on these questions, and social and politi-
cal values weigh in heavily. In a recent long-range strategic planning docu-
ment, EPA identified the “…need to develop the scientific understanding
and tools to better measure, model, and maintain or restore the integrity
and sustainability of ecosystems at local, regional, and national scales now
and in the future…” as a primary target for its near-term research programs.
29
5)RISK
Engaging the Public
in Decisions
30
working to address. Even accounting for this factor, it is clear from survey
data that a sizable gap exists between risk experts and nontechnical citi-
zens over how to appropriately define and measure risk.
Risk experts and officials with technical training tend to focus on—and
are comfortable with—the standard concept of risk as the possibility of
damage or unwanted effects. By contrast, the nontechnical public often
expands the concept of risk to include various nondamage attributes (see
Table 2). These added considerations are diverse, but in a general way
reflect both societal values and the play of the anger and fear that hazards
can readily evoke.
Such considerations can be extremely influential in shaping the public’s
reaction to a hazard—even to the point of overpowering scientific findings
about the magnitude of the risk. For example, pesticide residues in food
and the operation of nuclear power plants continue to be very prominent
public concerns—both of which scientific risk calculations indicate to be
relatively small hazards. Although scientific findings consider high-fat
diets or exposure to radon to be more serious public health concerns, they
draw far less public attention.
TABLE 2
Some Key Nondamage Attributes of Risk
ATTRIBUTE NATURE
Involuntary Risks voluntarily assumed are ranked differently from those
imposed by others.
Unfamiliar Generally speaking, more familiar risks are regarded as more acceptable.
Dreadful Risks that cause highly feared or dreaded consequences are viewed as
more dangerous.
chapter five
Catastrophic Large-scale disasters such as a plane crash weigh more seriously in the
public’s mind than individual events such as exposures to radon gas in a
neighbor’s basement.
Memorable Risks embedded in remarkable events have greater impact than risks that
arise in less prominent circumstances.
Unfair Substantial outrage is a more likely result if people feel they are being
wrongfully exposed.
Untrustworthy The level of outrage is higher if the source of the risk is not trusted.
Source: Foundation for American Communications and National Sea Grant College Program.
Reporting on Risk: A Handbook for Journalists and Citizens; The Annapolis Center, 1995, pp
84–86.
31
This gap between the public’s understanding of risk and that of risk
assessors is an obstacle both in making policy decisions about acceptable
risk and in effectively communicating the extent of risk that exists to the
public and decision makers.
32
6)RISKS
Evolving Efforts
To Compare and Rank
33
the EPA has led to a good many subsequent efforts. Several dozen com-
parative risk projects have been initiated by state governments and some
localities. Additionally, EPA has established information clearinghouses
and other programs to support this kind of effort.
Challenges in Conducting
Comparative RISK Analysis (CRA)
Currently, CRA consists of a pair of distinct analytical activities: (1) specif-
ic risk comparisons, which involve side-by-side evaluations of distinct risks,
on the basis of likelihood and severity of effects; and (2) programmatic
comparative risk assessment, which seeks to make “big picture” compar-
isons among many and widely differing hazards.
Specific risk comparisons can be particularly useful when one is consider-
ing the relative importance of risks within the context of similar products,
activities, or risk management actions. One application has been in the
area of risk communication, where such comparisons have been helpful in
facilitating nontechnical audiences’ understanding of the significance of
varying risk levels (for example, weighing the expected risks of new prod-
ucts or technologies against those that are already accepted or tolerated).
Paired comparisons of reasonably similar risks represent the most
straightforward application of CRA. Such evaluations might be conduct-
ed simply, based on estimated risk levels and the extent of anticipated
harm. For example, a pair of chemical pesticides might be compared with
respect to their expected chronic health effects, adjusted for likelihood.
Alternatively, one might compare the likelihood of serious injury posed by
driving to a desired destination against that of taking the train.
Even so, complexity can ratchet up quickly even with these simple
kinds of comparisons. The health effects relevant for consideration may
be multiple (e.g., cancer, reproductive effects, and organ dysfunction) and
may differ significantly by chemical (e.g., cancer predominates as the most
serious effect in one, but another health problem is primary in the other).
Additionally, it may also be appropriate to include adjustments for com-
pensating benefits (e.g., one of the chemicals may enhance the productivi-
ty of agricultural operations, whereas the other does not).
In general, the comparison process becomes more difficult as the dis-
similarity of the compared risks increases—indeed, to such a point that
the meaningfulness of the comparative exercise may be problematic. As
Table 3 outlines, differing risks can, and often do, take on disparate char-
acteristics—with respect to the source, kinds of effects, distribution over
space and time, and the nondamage attributes that may be relevant.
Furthermore, significant differences may exist with respect to the state of
34
knowledge about cause and effect (e.g., the risk of cancer from exposure
to a particular chemical may be poorly understood and take considerable
scientific research to resolve, whereas the risk factors in automobile acci-
dents are known with greater certainty).
TABLE 3
Dimensions in Which Risks Can Significantly Differ
FEATURES EXAMPLES
Source: Based on Office of Science and Technology Policy, Executive Office of the President.
Science, Risk, and Public Policy, March 1995.
35
in an agency’s portfolio. With programmatic CRA, much remains to be
done to develop well-defined guidance on how these studies should be
undertaken. All such projects face numerous questions at the outset that
go to the heart of what the project will be (see Table 4). Many of these
issues come with various options and involve nontrivial considerations
about analytical methods and limitations, study process, and the scope and
depth of the desired findings.
TABLE 4
Major Issues in Designing a CRA Project
Follow-up applications of What steps will be taken with the study’s findings to
study findings raise the level of understanding about significant
hazards, improve communications among those
involved in the policy process, and build consensus
toward appropriate action?
Source: Davies, J. Clarence, Ed., Comparing Environmental Risks: Tools for Setting Government
Priorities, Resources for the Future: Washington, DC, 1996.
36
in gauging the seriousness of a hazard and in establishing priorities.
Arguably, the strength of programmatic CRA is the opportunity it pro-
vides for discussion and debate among various important points of view:
technical experts, policy makers, and the public. Additionally, the process
facilitates addressing important public policy questions, which, although
difficult, are essential to address and which will be resolved through other,
less systematic approaches if not answered through CRA.
chapter six
37
FURTHER READING
PERSPECTIVES ON THE RISK ANALYSIS PROCESS
National Research Council, Committee on Risk Characterization.
Understanding Risk: Informing Decisions in a Democratic Society;
National Academy Press: Washington, DC, 1996.
38
DC, November 1993.
Kunreuther, Howard; Slovic, Paul. Science, Values, and Risk. Annals of the
American Academy of Political and Social Science, 1996, 545, 116–25.
39
Printed on recycled paper